4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
78 #include <asm/pgtable.h>
79 #include <asm/pgalloc.h>
80 #include <asm/uaccess.h>
81 #include <asm/mmu_context.h>
82 #include <asm/cacheflush.h>
83 #include <asm/tlbflush.h>
85 #include <trace/events/sched.h>
87 #define CREATE_TRACE_POINTS
88 #include <trace/events/task.h>
91 * Protected counters by write_lock_irq(&tasklist_lock)
93 unsigned long total_forks; /* Handle normal Linux uptimes. */
94 int nr_threads; /* The idle threads do not count.. */
96 int max_threads; /* tunable limit on nr_threads */
98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
102 #ifdef CONFIG_PROVE_RCU
103 int lockdep_tasklist_lock_is_held(void)
105 return lockdep_is_held(&tasklist_lock);
107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
108 #endif /* #ifdef CONFIG_PROVE_RCU */
110 int nr_processes(void)
115 for_each_possible_cpu(cpu)
116 total += per_cpu(process_counts, cpu);
121 void __weak arch_release_task_struct(struct task_struct *tsk)
125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
126 static struct kmem_cache *task_struct_cachep;
128 static inline struct task_struct *alloc_task_struct_node(int node)
130 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
133 static inline void free_task_struct(struct task_struct *tsk)
135 kmem_cache_free(task_struct_cachep, tsk);
139 void __weak arch_release_thread_info(struct thread_info *ti)
143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
146 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
147 * kmemcache based allocator.
149 # if THREAD_SIZE >= PAGE_SIZE
150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
153 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
156 return page ? page_address(page) : NULL;
159 static inline void free_thread_info(struct thread_info *ti)
161 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
164 static struct kmem_cache *thread_info_cache;
166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
169 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
172 static void free_thread_info(struct thread_info *ti)
174 kmem_cache_free(thread_info_cache, ti);
177 void thread_info_cache_init(void)
179 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
180 THREAD_SIZE, 0, NULL);
181 BUG_ON(thread_info_cache == NULL);
186 /* SLAB cache for signal_struct structures (tsk->signal) */
187 static struct kmem_cache *signal_cachep;
189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
190 struct kmem_cache *sighand_cachep;
192 /* SLAB cache for files_struct structures (tsk->files) */
193 struct kmem_cache *files_cachep;
195 /* SLAB cache for fs_struct structures (tsk->fs) */
196 struct kmem_cache *fs_cachep;
198 /* SLAB cache for vm_area_struct structures */
199 struct kmem_cache *vm_area_cachep;
201 /* SLAB cache for mm_struct structures (tsk->mm) */
202 static struct kmem_cache *mm_cachep;
204 static void account_kernel_stack(struct thread_info *ti, int account)
206 struct zone *zone = page_zone(virt_to_page(ti));
208 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
211 void free_task(struct task_struct *tsk)
213 account_kernel_stack(tsk->stack, -1);
214 arch_release_thread_info(tsk->stack);
215 free_thread_info(tsk->stack);
216 rt_mutex_debug_task_free(tsk);
217 ftrace_graph_exit_task(tsk);
218 put_seccomp_filter(tsk);
219 arch_release_task_struct(tsk);
220 free_task_struct(tsk);
222 EXPORT_SYMBOL(free_task);
224 static inline void free_signal_struct(struct signal_struct *sig)
226 taskstats_tgid_free(sig);
227 sched_autogroup_exit(sig);
228 kmem_cache_free(signal_cachep, sig);
231 static inline void put_signal_struct(struct signal_struct *sig)
233 if (atomic_dec_and_test(&sig->sigcnt))
234 free_signal_struct(sig);
237 void __put_task_struct(struct task_struct *tsk)
239 WARN_ON(!tsk->exit_state);
240 WARN_ON(atomic_read(&tsk->usage));
241 WARN_ON(tsk == current);
244 security_task_free(tsk);
246 delayacct_tsk_free(tsk);
247 put_signal_struct(tsk->signal);
249 if (!profile_handoff_task(tsk))
252 EXPORT_SYMBOL_GPL(__put_task_struct);
254 void __init __weak arch_task_cache_init(void) { }
259 static void set_max_threads(void)
262 * The default maximum number of threads is set to a safe
263 * value: the thread structures can take up at most one
264 * eighth of the memory.
266 max_threads = totalram_pages / (8 * THREAD_SIZE / PAGE_SIZE);
269 * we need to allow at least 20 threads to boot a system
271 if (max_threads < 20)
275 void __init fork_init(void)
277 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
278 #ifndef ARCH_MIN_TASKALIGN
279 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
281 /* create a slab on which task_structs can be allocated */
283 kmem_cache_create("task_struct", sizeof(struct task_struct),
284 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
287 /* do the arch specific task caches init */
288 arch_task_cache_init();
292 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
293 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
294 init_task.signal->rlim[RLIMIT_SIGPENDING] =
295 init_task.signal->rlim[RLIMIT_NPROC];
298 int __weak arch_dup_task_struct(struct task_struct *dst,
299 struct task_struct *src)
305 void set_task_stack_end_magic(struct task_struct *tsk)
307 unsigned long *stackend;
309 stackend = end_of_stack(tsk);
310 *stackend = STACK_END_MAGIC; /* for overflow detection */
313 static struct task_struct *dup_task_struct(struct task_struct *orig)
315 struct task_struct *tsk;
316 struct thread_info *ti;
317 int node = tsk_fork_get_node(orig);
320 tsk = alloc_task_struct_node(node);
324 ti = alloc_thread_info_node(tsk, node);
328 err = arch_dup_task_struct(tsk, orig);
333 #ifdef CONFIG_SECCOMP
335 * We must handle setting up seccomp filters once we're under
336 * the sighand lock in case orig has changed between now and
337 * then. Until then, filter must be NULL to avoid messing up
338 * the usage counts on the error path calling free_task.
340 tsk->seccomp.filter = NULL;
343 setup_thread_stack(tsk, orig);
344 clear_user_return_notifier(tsk);
345 clear_tsk_need_resched(tsk);
346 set_task_stack_end_magic(tsk);
348 #ifdef CONFIG_CC_STACKPROTECTOR
349 tsk->stack_canary = get_random_int();
353 * One for us, one for whoever does the "release_task()" (usually
356 atomic_set(&tsk->usage, 2);
357 #ifdef CONFIG_BLK_DEV_IO_TRACE
360 tsk->splice_pipe = NULL;
361 tsk->task_frag.page = NULL;
363 account_kernel_stack(ti, 1);
368 free_thread_info(ti);
370 free_task_struct(tsk);
375 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
377 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
378 struct rb_node **rb_link, *rb_parent;
380 unsigned long charge;
382 uprobe_start_dup_mmap();
383 down_write(&oldmm->mmap_sem);
384 flush_cache_dup_mm(oldmm);
385 uprobe_dup_mmap(oldmm, mm);
387 * Not linked in yet - no deadlock potential:
389 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
391 mm->total_vm = oldmm->total_vm;
392 mm->shared_vm = oldmm->shared_vm;
393 mm->exec_vm = oldmm->exec_vm;
394 mm->stack_vm = oldmm->stack_vm;
396 rb_link = &mm->mm_rb.rb_node;
399 retval = ksm_fork(mm, oldmm);
402 retval = khugepaged_fork(mm, oldmm);
407 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
410 if (mpnt->vm_flags & VM_DONTCOPY) {
411 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
416 if (mpnt->vm_flags & VM_ACCOUNT) {
417 unsigned long len = vma_pages(mpnt);
419 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
423 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
427 INIT_LIST_HEAD(&tmp->anon_vma_chain);
428 retval = vma_dup_policy(mpnt, tmp);
430 goto fail_nomem_policy;
432 if (anon_vma_fork(tmp, mpnt))
433 goto fail_nomem_anon_vma_fork;
434 tmp->vm_flags &= ~VM_LOCKED;
435 tmp->vm_next = tmp->vm_prev = NULL;
438 struct inode *inode = file_inode(file);
439 struct address_space *mapping = file->f_mapping;
442 if (tmp->vm_flags & VM_DENYWRITE)
443 atomic_dec(&inode->i_writecount);
444 i_mmap_lock_write(mapping);
445 if (tmp->vm_flags & VM_SHARED)
446 atomic_inc(&mapping->i_mmap_writable);
447 flush_dcache_mmap_lock(mapping);
448 /* insert tmp into the share list, just after mpnt */
449 vma_interval_tree_insert_after(tmp, mpnt,
451 flush_dcache_mmap_unlock(mapping);
452 i_mmap_unlock_write(mapping);
456 * Clear hugetlb-related page reserves for children. This only
457 * affects MAP_PRIVATE mappings. Faults generated by the child
458 * are not guaranteed to succeed, even if read-only
460 if (is_vm_hugetlb_page(tmp))
461 reset_vma_resv_huge_pages(tmp);
464 * Link in the new vma and copy the page table entries.
467 pprev = &tmp->vm_next;
471 __vma_link_rb(mm, tmp, rb_link, rb_parent);
472 rb_link = &tmp->vm_rb.rb_right;
473 rb_parent = &tmp->vm_rb;
476 retval = copy_page_range(mm, oldmm, mpnt);
478 if (tmp->vm_ops && tmp->vm_ops->open)
479 tmp->vm_ops->open(tmp);
484 /* a new mm has just been created */
485 arch_dup_mmap(oldmm, mm);
488 up_write(&mm->mmap_sem);
490 up_write(&oldmm->mmap_sem);
491 uprobe_end_dup_mmap();
493 fail_nomem_anon_vma_fork:
494 mpol_put(vma_policy(tmp));
496 kmem_cache_free(vm_area_cachep, tmp);
499 vm_unacct_memory(charge);
503 static inline int mm_alloc_pgd(struct mm_struct *mm)
505 mm->pgd = pgd_alloc(mm);
506 if (unlikely(!mm->pgd))
511 static inline void mm_free_pgd(struct mm_struct *mm)
513 pgd_free(mm, mm->pgd);
516 #define dup_mmap(mm, oldmm) (0)
517 #define mm_alloc_pgd(mm) (0)
518 #define mm_free_pgd(mm)
519 #endif /* CONFIG_MMU */
521 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
523 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
524 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
526 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
528 static int __init coredump_filter_setup(char *s)
530 default_dump_filter =
531 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
532 MMF_DUMP_FILTER_MASK;
536 __setup("coredump_filter=", coredump_filter_setup);
538 #include <linux/init_task.h>
540 static void mm_init_aio(struct mm_struct *mm)
543 spin_lock_init(&mm->ioctx_lock);
544 mm->ioctx_table = NULL;
548 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
555 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
559 mm->vmacache_seqnum = 0;
560 atomic_set(&mm->mm_users, 1);
561 atomic_set(&mm->mm_count, 1);
562 init_rwsem(&mm->mmap_sem);
563 INIT_LIST_HEAD(&mm->mmlist);
564 mm->core_state = NULL;
565 atomic_long_set(&mm->nr_ptes, 0);
570 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
571 spin_lock_init(&mm->page_table_lock);
574 mm_init_owner(mm, p);
575 mmu_notifier_mm_init(mm);
576 clear_tlb_flush_pending(mm);
577 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
578 mm->pmd_huge_pte = NULL;
582 mm->flags = current->mm->flags & MMF_INIT_MASK;
583 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
585 mm->flags = default_dump_filter;
589 if (mm_alloc_pgd(mm))
592 if (init_new_context(p, mm))
604 static void check_mm(struct mm_struct *mm)
608 for (i = 0; i < NR_MM_COUNTERS; i++) {
609 long x = atomic_long_read(&mm->rss_stat.count[i]);
612 printk(KERN_ALERT "BUG: Bad rss-counter state "
613 "mm:%p idx:%d val:%ld\n", mm, i, x);
616 if (atomic_long_read(&mm->nr_ptes))
617 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
618 atomic_long_read(&mm->nr_ptes));
620 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
624 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
629 * Allocate and initialize an mm_struct.
631 struct mm_struct *mm_alloc(void)
633 struct mm_struct *mm;
639 memset(mm, 0, sizeof(*mm));
640 return mm_init(mm, current);
644 * Called when the last reference to the mm
645 * is dropped: either by a lazy thread or by
646 * mmput. Free the page directory and the mm.
648 void __mmdrop(struct mm_struct *mm)
650 BUG_ON(mm == &init_mm);
653 mmu_notifier_mm_destroy(mm);
657 EXPORT_SYMBOL_GPL(__mmdrop);
660 * Decrement the use count and release all resources for an mm.
662 void mmput(struct mm_struct *mm)
666 if (atomic_dec_and_test(&mm->mm_users)) {
667 uprobe_clear_state(mm);
670 khugepaged_exit(mm); /* must run before exit_mmap */
672 set_mm_exe_file(mm, NULL);
673 if (!list_empty(&mm->mmlist)) {
674 spin_lock(&mmlist_lock);
675 list_del(&mm->mmlist);
676 spin_unlock(&mmlist_lock);
679 module_put(mm->binfmt->module);
683 EXPORT_SYMBOL_GPL(mmput);
685 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
688 get_file(new_exe_file);
691 mm->exe_file = new_exe_file;
694 struct file *get_mm_exe_file(struct mm_struct *mm)
696 struct file *exe_file;
698 /* We need mmap_sem to protect against races with removal of exe_file */
699 down_read(&mm->mmap_sem);
700 exe_file = mm->exe_file;
703 up_read(&mm->mmap_sem);
707 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
709 /* It's safe to write the exe_file pointer without exe_file_lock because
710 * this is called during fork when the task is not yet in /proc */
711 newmm->exe_file = get_mm_exe_file(oldmm);
715 * get_task_mm - acquire a reference to the task's mm
717 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
718 * this kernel workthread has transiently adopted a user mm with use_mm,
719 * to do its AIO) is not set and if so returns a reference to it, after
720 * bumping up the use count. User must release the mm via mmput()
721 * after use. Typically used by /proc and ptrace.
723 struct mm_struct *get_task_mm(struct task_struct *task)
725 struct mm_struct *mm;
730 if (task->flags & PF_KTHREAD)
733 atomic_inc(&mm->mm_users);
738 EXPORT_SYMBOL_GPL(get_task_mm);
740 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
742 struct mm_struct *mm;
745 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
749 mm = get_task_mm(task);
750 if (mm && mm != current->mm &&
751 !ptrace_may_access(task, mode)) {
753 mm = ERR_PTR(-EACCES);
755 mutex_unlock(&task->signal->cred_guard_mutex);
760 static void complete_vfork_done(struct task_struct *tsk)
762 struct completion *vfork;
765 vfork = tsk->vfork_done;
767 tsk->vfork_done = NULL;
773 static int wait_for_vfork_done(struct task_struct *child,
774 struct completion *vfork)
778 freezer_do_not_count();
779 killed = wait_for_completion_killable(vfork);
784 child->vfork_done = NULL;
788 put_task_struct(child);
792 /* Please note the differences between mmput and mm_release.
793 * mmput is called whenever we stop holding onto a mm_struct,
794 * error success whatever.
796 * mm_release is called after a mm_struct has been removed
797 * from the current process.
799 * This difference is important for error handling, when we
800 * only half set up a mm_struct for a new process and need to restore
801 * the old one. Because we mmput the new mm_struct before
802 * restoring the old one. . .
803 * Eric Biederman 10 January 1998
805 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
807 /* Get rid of any futexes when releasing the mm */
809 if (unlikely(tsk->robust_list)) {
810 exit_robust_list(tsk);
811 tsk->robust_list = NULL;
814 if (unlikely(tsk->compat_robust_list)) {
815 compat_exit_robust_list(tsk);
816 tsk->compat_robust_list = NULL;
819 if (unlikely(!list_empty(&tsk->pi_state_list)))
820 exit_pi_state_list(tsk);
823 uprobe_free_utask(tsk);
825 /* Get rid of any cached register state */
826 deactivate_mm(tsk, mm);
829 * If we're exiting normally, clear a user-space tid field if
830 * requested. We leave this alone when dying by signal, to leave
831 * the value intact in a core dump, and to save the unnecessary
832 * trouble, say, a killed vfork parent shouldn't touch this mm.
833 * Userland only wants this done for a sys_exit.
835 if (tsk->clear_child_tid) {
836 if (!(tsk->flags & PF_SIGNALED) &&
837 atomic_read(&mm->mm_users) > 1) {
839 * We don't check the error code - if userspace has
840 * not set up a proper pointer then tough luck.
842 put_user(0, tsk->clear_child_tid);
843 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
846 tsk->clear_child_tid = NULL;
850 * All done, finally we can wake up parent and return this mm to him.
851 * Also kthread_stop() uses this completion for synchronization.
854 complete_vfork_done(tsk);
858 * Allocate a new mm structure and copy contents from the
859 * mm structure of the passed in task structure.
861 static struct mm_struct *dup_mm(struct task_struct *tsk)
863 struct mm_struct *mm, *oldmm = current->mm;
870 memcpy(mm, oldmm, sizeof(*mm));
872 if (!mm_init(mm, tsk))
875 dup_mm_exe_file(oldmm, mm);
877 err = dup_mmap(mm, oldmm);
881 mm->hiwater_rss = get_mm_rss(mm);
882 mm->hiwater_vm = mm->total_vm;
884 if (mm->binfmt && !try_module_get(mm->binfmt->module))
890 /* don't put binfmt in mmput, we haven't got module yet */
898 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
900 struct mm_struct *mm, *oldmm;
903 tsk->min_flt = tsk->maj_flt = 0;
904 tsk->nvcsw = tsk->nivcsw = 0;
905 #ifdef CONFIG_DETECT_HUNG_TASK
906 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
910 tsk->active_mm = NULL;
913 * Are we cloning a kernel thread?
915 * We need to steal a active VM for that..
921 /* initialize the new vmacache entries */
924 if (clone_flags & CLONE_VM) {
925 atomic_inc(&oldmm->mm_users);
944 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
946 struct fs_struct *fs = current->fs;
947 if (clone_flags & CLONE_FS) {
948 /* tsk->fs is already what we want */
949 spin_lock(&fs->lock);
951 spin_unlock(&fs->lock);
955 spin_unlock(&fs->lock);
958 tsk->fs = copy_fs_struct(fs);
964 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
966 struct files_struct *oldf, *newf;
970 * A background process may not have any files ...
972 oldf = current->files;
976 if (clone_flags & CLONE_FILES) {
977 atomic_inc(&oldf->count);
981 newf = dup_fd(oldf, &error);
991 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
994 struct io_context *ioc = current->io_context;
995 struct io_context *new_ioc;
1000 * Share io context with parent, if CLONE_IO is set
1002 if (clone_flags & CLONE_IO) {
1004 tsk->io_context = ioc;
1005 } else if (ioprio_valid(ioc->ioprio)) {
1006 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1007 if (unlikely(!new_ioc))
1010 new_ioc->ioprio = ioc->ioprio;
1011 put_io_context(new_ioc);
1017 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1019 struct sighand_struct *sig;
1021 if (clone_flags & CLONE_SIGHAND) {
1022 atomic_inc(¤t->sighand->count);
1025 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1026 rcu_assign_pointer(tsk->sighand, sig);
1029 atomic_set(&sig->count, 1);
1030 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1034 void __cleanup_sighand(struct sighand_struct *sighand)
1036 if (atomic_dec_and_test(&sighand->count)) {
1037 signalfd_cleanup(sighand);
1039 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1040 * without an RCU grace period, see __lock_task_sighand().
1042 kmem_cache_free(sighand_cachep, sighand);
1047 * Initialize POSIX timer handling for a thread group.
1049 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1051 unsigned long cpu_limit;
1053 /* Thread group counters. */
1054 thread_group_cputime_init(sig);
1056 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1057 if (cpu_limit != RLIM_INFINITY) {
1058 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1059 sig->cputimer.running = 1;
1062 /* The timer lists. */
1063 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1064 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1065 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1068 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1070 struct signal_struct *sig;
1072 if (clone_flags & CLONE_THREAD)
1075 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1080 sig->nr_threads = 1;
1081 atomic_set(&sig->live, 1);
1082 atomic_set(&sig->sigcnt, 1);
1084 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1085 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1086 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1088 init_waitqueue_head(&sig->wait_chldexit);
1089 sig->curr_target = tsk;
1090 init_sigpending(&sig->shared_pending);
1091 INIT_LIST_HEAD(&sig->posix_timers);
1092 seqlock_init(&sig->stats_lock);
1094 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1095 sig->real_timer.function = it_real_fn;
1097 task_lock(current->group_leader);
1098 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1099 task_unlock(current->group_leader);
1101 posix_cpu_timers_init_group(sig);
1103 tty_audit_fork(sig);
1104 sched_autogroup_fork(sig);
1106 #ifdef CONFIG_CGROUPS
1107 init_rwsem(&sig->group_rwsem);
1110 sig->oom_score_adj = current->signal->oom_score_adj;
1111 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1113 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1114 current->signal->is_child_subreaper;
1116 mutex_init(&sig->cred_guard_mutex);
1121 static void copy_seccomp(struct task_struct *p)
1123 #ifdef CONFIG_SECCOMP
1125 * Must be called with sighand->lock held, which is common to
1126 * all threads in the group. Holding cred_guard_mutex is not
1127 * needed because this new task is not yet running and cannot
1130 assert_spin_locked(¤t->sighand->siglock);
1132 /* Ref-count the new filter user, and assign it. */
1133 get_seccomp_filter(current);
1134 p->seccomp = current->seccomp;
1137 * Explicitly enable no_new_privs here in case it got set
1138 * between the task_struct being duplicated and holding the
1139 * sighand lock. The seccomp state and nnp must be in sync.
1141 if (task_no_new_privs(current))
1142 task_set_no_new_privs(p);
1145 * If the parent gained a seccomp mode after copying thread
1146 * flags and between before we held the sighand lock, we have
1147 * to manually enable the seccomp thread flag here.
1149 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1150 set_tsk_thread_flag(p, TIF_SECCOMP);
1154 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1156 current->clear_child_tid = tidptr;
1158 return task_pid_vnr(current);
1161 static void rt_mutex_init_task(struct task_struct *p)
1163 raw_spin_lock_init(&p->pi_lock);
1164 #ifdef CONFIG_RT_MUTEXES
1165 p->pi_waiters = RB_ROOT;
1166 p->pi_waiters_leftmost = NULL;
1167 p->pi_blocked_on = NULL;
1172 * Initialize POSIX timer handling for a single task.
1174 static void posix_cpu_timers_init(struct task_struct *tsk)
1176 tsk->cputime_expires.prof_exp = 0;
1177 tsk->cputime_expires.virt_exp = 0;
1178 tsk->cputime_expires.sched_exp = 0;
1179 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1180 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1181 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1185 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1187 task->pids[type].pid = pid;
1191 * This creates a new process as a copy of the old one,
1192 * but does not actually start it yet.
1194 * It copies the registers, and all the appropriate
1195 * parts of the process environment (as per the clone
1196 * flags). The actual kick-off is left to the caller.
1198 static struct task_struct *copy_process(unsigned long clone_flags,
1199 unsigned long stack_start,
1200 unsigned long stack_size,
1201 int __user *child_tidptr,
1206 struct task_struct *p;
1208 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1209 return ERR_PTR(-EINVAL);
1211 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1212 return ERR_PTR(-EINVAL);
1215 * Thread groups must share signals as well, and detached threads
1216 * can only be started up within the thread group.
1218 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1219 return ERR_PTR(-EINVAL);
1222 * Shared signal handlers imply shared VM. By way of the above,
1223 * thread groups also imply shared VM. Blocking this case allows
1224 * for various simplifications in other code.
1226 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1227 return ERR_PTR(-EINVAL);
1230 * Siblings of global init remain as zombies on exit since they are
1231 * not reaped by their parent (swapper). To solve this and to avoid
1232 * multi-rooted process trees, prevent global and container-inits
1233 * from creating siblings.
1235 if ((clone_flags & CLONE_PARENT) &&
1236 current->signal->flags & SIGNAL_UNKILLABLE)
1237 return ERR_PTR(-EINVAL);
1240 * If the new process will be in a different pid or user namespace
1241 * do not allow it to share a thread group or signal handlers or
1242 * parent with the forking task.
1244 if (clone_flags & CLONE_SIGHAND) {
1245 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1246 (task_active_pid_ns(current) !=
1247 current->nsproxy->pid_ns_for_children))
1248 return ERR_PTR(-EINVAL);
1251 retval = security_task_create(clone_flags);
1256 p = dup_task_struct(current);
1260 ftrace_graph_init_task(p);
1262 rt_mutex_init_task(p);
1264 #ifdef CONFIG_PROVE_LOCKING
1265 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1266 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1269 if (atomic_read(&p->real_cred->user->processes) >=
1270 task_rlimit(p, RLIMIT_NPROC)) {
1271 if (p->real_cred->user != INIT_USER &&
1272 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1275 current->flags &= ~PF_NPROC_EXCEEDED;
1277 retval = copy_creds(p, clone_flags);
1282 * If multiple threads are within copy_process(), then this check
1283 * triggers too late. This doesn't hurt, the check is only there
1284 * to stop root fork bombs.
1287 if (nr_threads >= max_threads)
1288 goto bad_fork_cleanup_count;
1290 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1291 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1292 p->flags |= PF_FORKNOEXEC;
1293 INIT_LIST_HEAD(&p->children);
1294 INIT_LIST_HEAD(&p->sibling);
1295 rcu_copy_process(p);
1296 p->vfork_done = NULL;
1297 spin_lock_init(&p->alloc_lock);
1299 init_sigpending(&p->pending);
1301 p->utime = p->stime = p->gtime = 0;
1302 p->utimescaled = p->stimescaled = 0;
1303 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1304 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1306 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1307 seqlock_init(&p->vtime_seqlock);
1309 p->vtime_snap_whence = VTIME_SLEEPING;
1312 #if defined(SPLIT_RSS_COUNTING)
1313 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1316 p->default_timer_slack_ns = current->timer_slack_ns;
1318 task_io_accounting_init(&p->ioac);
1319 acct_clear_integrals(p);
1321 posix_cpu_timers_init(p);
1323 p->start_time = ktime_get_ns();
1324 p->real_start_time = ktime_get_boot_ns();
1325 p->io_context = NULL;
1326 p->audit_context = NULL;
1327 if (clone_flags & CLONE_THREAD)
1328 threadgroup_change_begin(current);
1331 p->mempolicy = mpol_dup(p->mempolicy);
1332 if (IS_ERR(p->mempolicy)) {
1333 retval = PTR_ERR(p->mempolicy);
1334 p->mempolicy = NULL;
1335 goto bad_fork_cleanup_threadgroup_lock;
1338 #ifdef CONFIG_CPUSETS
1339 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1340 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1341 seqcount_init(&p->mems_allowed_seq);
1343 #ifdef CONFIG_TRACE_IRQFLAGS
1345 p->hardirqs_enabled = 0;
1346 p->hardirq_enable_ip = 0;
1347 p->hardirq_enable_event = 0;
1348 p->hardirq_disable_ip = _THIS_IP_;
1349 p->hardirq_disable_event = 0;
1350 p->softirqs_enabled = 1;
1351 p->softirq_enable_ip = _THIS_IP_;
1352 p->softirq_enable_event = 0;
1353 p->softirq_disable_ip = 0;
1354 p->softirq_disable_event = 0;
1355 p->hardirq_context = 0;
1356 p->softirq_context = 0;
1358 #ifdef CONFIG_LOCKDEP
1359 p->lockdep_depth = 0; /* no locks held yet */
1360 p->curr_chain_key = 0;
1361 p->lockdep_recursion = 0;
1364 #ifdef CONFIG_DEBUG_MUTEXES
1365 p->blocked_on = NULL; /* not blocked yet */
1367 #ifdef CONFIG_BCACHE
1368 p->sequential_io = 0;
1369 p->sequential_io_avg = 0;
1372 /* Perform scheduler related setup. Assign this task to a CPU. */
1373 retval = sched_fork(clone_flags, p);
1375 goto bad_fork_cleanup_policy;
1377 retval = perf_event_init_task(p);
1379 goto bad_fork_cleanup_policy;
1380 retval = audit_alloc(p);
1382 goto bad_fork_cleanup_perf;
1383 /* copy all the process information */
1385 retval = copy_semundo(clone_flags, p);
1387 goto bad_fork_cleanup_audit;
1388 retval = copy_files(clone_flags, p);
1390 goto bad_fork_cleanup_semundo;
1391 retval = copy_fs(clone_flags, p);
1393 goto bad_fork_cleanup_files;
1394 retval = copy_sighand(clone_flags, p);
1396 goto bad_fork_cleanup_fs;
1397 retval = copy_signal(clone_flags, p);
1399 goto bad_fork_cleanup_sighand;
1400 retval = copy_mm(clone_flags, p);
1402 goto bad_fork_cleanup_signal;
1403 retval = copy_namespaces(clone_flags, p);
1405 goto bad_fork_cleanup_mm;
1406 retval = copy_io(clone_flags, p);
1408 goto bad_fork_cleanup_namespaces;
1409 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1411 goto bad_fork_cleanup_io;
1413 if (pid != &init_struct_pid) {
1414 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1416 retval = PTR_ERR(pid);
1417 goto bad_fork_cleanup_io;
1421 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1423 * Clear TID on mm_release()?
1425 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1430 p->robust_list = NULL;
1431 #ifdef CONFIG_COMPAT
1432 p->compat_robust_list = NULL;
1434 INIT_LIST_HEAD(&p->pi_state_list);
1435 p->pi_state_cache = NULL;
1438 * sigaltstack should be cleared when sharing the same VM
1440 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1441 p->sas_ss_sp = p->sas_ss_size = 0;
1444 * Syscall tracing and stepping should be turned off in the
1445 * child regardless of CLONE_PTRACE.
1447 user_disable_single_step(p);
1448 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1449 #ifdef TIF_SYSCALL_EMU
1450 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1452 clear_all_latency_tracing(p);
1454 /* ok, now we should be set up.. */
1455 p->pid = pid_nr(pid);
1456 if (clone_flags & CLONE_THREAD) {
1457 p->exit_signal = -1;
1458 p->group_leader = current->group_leader;
1459 p->tgid = current->tgid;
1461 if (clone_flags & CLONE_PARENT)
1462 p->exit_signal = current->group_leader->exit_signal;
1464 p->exit_signal = (clone_flags & CSIGNAL);
1465 p->group_leader = p;
1470 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1471 p->dirty_paused_when = 0;
1473 p->pdeath_signal = 0;
1474 INIT_LIST_HEAD(&p->thread_group);
1475 p->task_works = NULL;
1478 * Make it visible to the rest of the system, but dont wake it up yet.
1479 * Need tasklist lock for parent etc handling!
1481 write_lock_irq(&tasklist_lock);
1483 /* CLONE_PARENT re-uses the old parent */
1484 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1485 p->real_parent = current->real_parent;
1486 p->parent_exec_id = current->parent_exec_id;
1488 p->real_parent = current;
1489 p->parent_exec_id = current->self_exec_id;
1492 spin_lock(¤t->sighand->siglock);
1495 * Copy seccomp details explicitly here, in case they were changed
1496 * before holding sighand lock.
1501 * Process group and session signals need to be delivered to just the
1502 * parent before the fork or both the parent and the child after the
1503 * fork. Restart if a signal comes in before we add the new process to
1504 * it's process group.
1505 * A fatal signal pending means that current will exit, so the new
1506 * thread can't slip out of an OOM kill (or normal SIGKILL).
1508 recalc_sigpending();
1509 if (signal_pending(current)) {
1510 spin_unlock(¤t->sighand->siglock);
1511 write_unlock_irq(&tasklist_lock);
1512 retval = -ERESTARTNOINTR;
1513 goto bad_fork_free_pid;
1516 if (likely(p->pid)) {
1517 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1519 init_task_pid(p, PIDTYPE_PID, pid);
1520 if (thread_group_leader(p)) {
1521 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1522 init_task_pid(p, PIDTYPE_SID, task_session(current));
1524 if (is_child_reaper(pid)) {
1525 ns_of_pid(pid)->child_reaper = p;
1526 p->signal->flags |= SIGNAL_UNKILLABLE;
1529 p->signal->leader_pid = pid;
1530 p->signal->tty = tty_kref_get(current->signal->tty);
1531 list_add_tail(&p->sibling, &p->real_parent->children);
1532 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1533 attach_pid(p, PIDTYPE_PGID);
1534 attach_pid(p, PIDTYPE_SID);
1535 __this_cpu_inc(process_counts);
1537 current->signal->nr_threads++;
1538 atomic_inc(¤t->signal->live);
1539 atomic_inc(¤t->signal->sigcnt);
1540 list_add_tail_rcu(&p->thread_group,
1541 &p->group_leader->thread_group);
1542 list_add_tail_rcu(&p->thread_node,
1543 &p->signal->thread_head);
1545 attach_pid(p, PIDTYPE_PID);
1550 spin_unlock(¤t->sighand->siglock);
1551 syscall_tracepoint_update(p);
1552 write_unlock_irq(&tasklist_lock);
1554 proc_fork_connector(p);
1555 cgroup_post_fork(p);
1556 if (clone_flags & CLONE_THREAD)
1557 threadgroup_change_end(current);
1560 trace_task_newtask(p, clone_flags);
1561 uprobe_copy_process(p, clone_flags);
1566 if (pid != &init_struct_pid)
1568 bad_fork_cleanup_io:
1571 bad_fork_cleanup_namespaces:
1572 exit_task_namespaces(p);
1573 bad_fork_cleanup_mm:
1576 bad_fork_cleanup_signal:
1577 if (!(clone_flags & CLONE_THREAD))
1578 free_signal_struct(p->signal);
1579 bad_fork_cleanup_sighand:
1580 __cleanup_sighand(p->sighand);
1581 bad_fork_cleanup_fs:
1582 exit_fs(p); /* blocking */
1583 bad_fork_cleanup_files:
1584 exit_files(p); /* blocking */
1585 bad_fork_cleanup_semundo:
1587 bad_fork_cleanup_audit:
1589 bad_fork_cleanup_perf:
1590 perf_event_free_task(p);
1591 bad_fork_cleanup_policy:
1593 mpol_put(p->mempolicy);
1594 bad_fork_cleanup_threadgroup_lock:
1596 if (clone_flags & CLONE_THREAD)
1597 threadgroup_change_end(current);
1598 delayacct_tsk_free(p);
1599 bad_fork_cleanup_count:
1600 atomic_dec(&p->cred->user->processes);
1605 return ERR_PTR(retval);
1608 static inline void init_idle_pids(struct pid_link *links)
1612 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1613 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1614 links[type].pid = &init_struct_pid;
1618 struct task_struct *fork_idle(int cpu)
1620 struct task_struct *task;
1621 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1622 if (!IS_ERR(task)) {
1623 init_idle_pids(task->pids);
1624 init_idle(task, cpu);
1631 * Ok, this is the main fork-routine.
1633 * It copies the process, and if successful kick-starts
1634 * it and waits for it to finish using the VM if required.
1636 long do_fork(unsigned long clone_flags,
1637 unsigned long stack_start,
1638 unsigned long stack_size,
1639 int __user *parent_tidptr,
1640 int __user *child_tidptr)
1642 struct task_struct *p;
1647 * Determine whether and which event to report to ptracer. When
1648 * called from kernel_thread or CLONE_UNTRACED is explicitly
1649 * requested, no event is reported; otherwise, report if the event
1650 * for the type of forking is enabled.
1652 if (!(clone_flags & CLONE_UNTRACED)) {
1653 if (clone_flags & CLONE_VFORK)
1654 trace = PTRACE_EVENT_VFORK;
1655 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1656 trace = PTRACE_EVENT_CLONE;
1658 trace = PTRACE_EVENT_FORK;
1660 if (likely(!ptrace_event_enabled(current, trace)))
1664 p = copy_process(clone_flags, stack_start, stack_size,
1665 child_tidptr, NULL, trace);
1667 * Do this prior waking up the new thread - the thread pointer
1668 * might get invalid after that point, if the thread exits quickly.
1671 struct completion vfork;
1674 trace_sched_process_fork(current, p);
1676 pid = get_task_pid(p, PIDTYPE_PID);
1679 if (clone_flags & CLONE_PARENT_SETTID)
1680 put_user(nr, parent_tidptr);
1682 if (clone_flags & CLONE_VFORK) {
1683 p->vfork_done = &vfork;
1684 init_completion(&vfork);
1688 wake_up_new_task(p);
1690 /* forking complete and child started to run, tell ptracer */
1691 if (unlikely(trace))
1692 ptrace_event_pid(trace, pid);
1694 if (clone_flags & CLONE_VFORK) {
1695 if (!wait_for_vfork_done(p, &vfork))
1696 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1707 * Create a kernel thread.
1709 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1711 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1712 (unsigned long)arg, NULL, NULL);
1715 #ifdef __ARCH_WANT_SYS_FORK
1716 SYSCALL_DEFINE0(fork)
1719 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1721 /* can not support in nommu mode */
1727 #ifdef __ARCH_WANT_SYS_VFORK
1728 SYSCALL_DEFINE0(vfork)
1730 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1735 #ifdef __ARCH_WANT_SYS_CLONE
1736 #ifdef CONFIG_CLONE_BACKWARDS
1737 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1738 int __user *, parent_tidptr,
1740 int __user *, child_tidptr)
1741 #elif defined(CONFIG_CLONE_BACKWARDS2)
1742 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1743 int __user *, parent_tidptr,
1744 int __user *, child_tidptr,
1746 #elif defined(CONFIG_CLONE_BACKWARDS3)
1747 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1749 int __user *, parent_tidptr,
1750 int __user *, child_tidptr,
1753 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1754 int __user *, parent_tidptr,
1755 int __user *, child_tidptr,
1759 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1763 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1764 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1767 static void sighand_ctor(void *data)
1769 struct sighand_struct *sighand = data;
1771 spin_lock_init(&sighand->siglock);
1772 init_waitqueue_head(&sighand->signalfd_wqh);
1775 void __init proc_caches_init(void)
1777 sighand_cachep = kmem_cache_create("sighand_cache",
1778 sizeof(struct sighand_struct), 0,
1779 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1780 SLAB_NOTRACK, sighand_ctor);
1781 signal_cachep = kmem_cache_create("signal_cache",
1782 sizeof(struct signal_struct), 0,
1783 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1784 files_cachep = kmem_cache_create("files_cache",
1785 sizeof(struct files_struct), 0,
1786 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1787 fs_cachep = kmem_cache_create("fs_cache",
1788 sizeof(struct fs_struct), 0,
1789 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1791 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1792 * whole struct cpumask for the OFFSTACK case. We could change
1793 * this to *only* allocate as much of it as required by the
1794 * maximum number of CPU's we can ever have. The cpumask_allocation
1795 * is at the end of the structure, exactly for that reason.
1797 mm_cachep = kmem_cache_create("mm_struct",
1798 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1799 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1800 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1802 nsproxy_cache_init();
1806 * Check constraints on flags passed to the unshare system call.
1808 static int check_unshare_flags(unsigned long unshare_flags)
1810 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1811 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1812 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1813 CLONE_NEWUSER|CLONE_NEWPID))
1816 * Not implemented, but pretend it works if there is nothing to
1817 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1818 * needs to unshare vm.
1820 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1821 /* FIXME: get_task_mm() increments ->mm_users */
1822 if (atomic_read(¤t->mm->mm_users) > 1)
1830 * Unshare the filesystem structure if it is being shared
1832 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1834 struct fs_struct *fs = current->fs;
1836 if (!(unshare_flags & CLONE_FS) || !fs)
1839 /* don't need lock here; in the worst case we'll do useless copy */
1843 *new_fsp = copy_fs_struct(fs);
1851 * Unshare file descriptor table if it is being shared
1853 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1855 struct files_struct *fd = current->files;
1858 if ((unshare_flags & CLONE_FILES) &&
1859 (fd && atomic_read(&fd->count) > 1)) {
1860 *new_fdp = dup_fd(fd, &error);
1869 * unshare allows a process to 'unshare' part of the process
1870 * context which was originally shared using clone. copy_*
1871 * functions used by do_fork() cannot be used here directly
1872 * because they modify an inactive task_struct that is being
1873 * constructed. Here we are modifying the current, active,
1876 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1878 struct fs_struct *fs, *new_fs = NULL;
1879 struct files_struct *fd, *new_fd = NULL;
1880 struct cred *new_cred = NULL;
1881 struct nsproxy *new_nsproxy = NULL;
1886 * If unsharing a user namespace must also unshare the thread.
1888 if (unshare_flags & CLONE_NEWUSER)
1889 unshare_flags |= CLONE_THREAD | CLONE_FS;
1891 * If unsharing a thread from a thread group, must also unshare vm.
1893 if (unshare_flags & CLONE_THREAD)
1894 unshare_flags |= CLONE_VM;
1896 * If unsharing vm, must also unshare signal handlers.
1898 if (unshare_flags & CLONE_VM)
1899 unshare_flags |= CLONE_SIGHAND;
1901 * If unsharing namespace, must also unshare filesystem information.
1903 if (unshare_flags & CLONE_NEWNS)
1904 unshare_flags |= CLONE_FS;
1906 err = check_unshare_flags(unshare_flags);
1908 goto bad_unshare_out;
1910 * CLONE_NEWIPC must also detach from the undolist: after switching
1911 * to a new ipc namespace, the semaphore arrays from the old
1912 * namespace are unreachable.
1914 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1916 err = unshare_fs(unshare_flags, &new_fs);
1918 goto bad_unshare_out;
1919 err = unshare_fd(unshare_flags, &new_fd);
1921 goto bad_unshare_cleanup_fs;
1922 err = unshare_userns(unshare_flags, &new_cred);
1924 goto bad_unshare_cleanup_fd;
1925 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1928 goto bad_unshare_cleanup_cred;
1930 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1933 * CLONE_SYSVSEM is equivalent to sys_exit().
1937 if (unshare_flags & CLONE_NEWIPC) {
1938 /* Orphan segments in old ns (see sem above). */
1940 shm_init_task(current);
1944 switch_task_namespaces(current, new_nsproxy);
1950 spin_lock(&fs->lock);
1951 current->fs = new_fs;
1956 spin_unlock(&fs->lock);
1960 fd = current->files;
1961 current->files = new_fd;
1965 task_unlock(current);
1968 /* Install the new user namespace */
1969 commit_creds(new_cred);
1974 bad_unshare_cleanup_cred:
1977 bad_unshare_cleanup_fd:
1979 put_files_struct(new_fd);
1981 bad_unshare_cleanup_fs:
1983 free_fs_struct(new_fs);
1990 * Helper to unshare the files of the current task.
1991 * We don't want to expose copy_files internals to
1992 * the exec layer of the kernel.
1995 int unshare_files(struct files_struct **displaced)
1997 struct task_struct *task = current;
1998 struct files_struct *copy = NULL;
2001 error = unshare_fd(CLONE_FILES, ©);
2002 if (error || !copy) {
2006 *displaced = task->files;